#NephJC Chat
Tuesday Sep 19 2023 9 pm Eastern
Wednesday Sep 20 2023 9 pm IST
Nature 2023 Jul;619(7970):585-594. doi: 10.1038/s41586-023-05769-3. Epub 2023 Jul 19.
An atlas of healthy and injured cell states and niches in the human kidney
Blue B Lake # 1 2, Rajasree Menon # 3, Seth Winfree # 4, Qiwen Hu # 5, Ricardo Melo Ferreira # 6, Kian Kalhor # 1, Daria Barwinska 6, Edgar A Otto 7, Michael Ferkowicz 6, Dinh Diep 1 2, Nongluk Plongthongkum 1, Amanda Knoten 8, Sarah Urata 1, Laura H Mariani 7, Abhijit S Naik 7, Sean Eddy 7, Bo Zhang 8, Yan Wu 1 2, Diane Salamon 8, James C Williams 6, Xin Wang 5, Karol S Balderrama 9, Paul J Hoover 9, Evan Murray 9, Jamie L Marshall 9, Teia Noel 9, Anitha Vijayan 8, Austin Hartman 10, Fei Chen 9, Sushrut S Waikar 11, Sylvia E Rosas 12 13, Francis P Wilson 14, Paul M Palevsky 15, Krzysztof Kiryluk 16, John R Sedor 17, Robert D Toto 18, Chirag R Parikh 19, Eric H Kim 20, Rahul Satija 10, Anna Greka 9, Evan Z Macosko 9, Peter V Kharchenko 5 2, Joseph P Gaut 21, Jeffrey B Hodgin 22; KPMP Consortium; Michael T Eadon 23, Pierre C Dagher 24, Tarek M El-Achkar 25, Kun Zhang 26 27, Matthias Kretzler 28, Sanjay Jain 29 30
PMID: 37468583
Introduction
Modern geographical maps are the remarkable outcome of centuries of collective effort, representing a continuous journey of refinement and innovation to achieve an ever-more precise depiction of our world. Similar efforts have been undertaken to understand the kidney, with several groups gleaning remarkable insights on a much shorter timescale. Decades of research have uncovered many of the kidney’s secrets, but the heterogeneity and diversity of kidney cell types in health and injury have remained challenging to fully elucidate using traditional laboratory techniques such as bulk RNA sequencing. Integrating several state-of-the-art single-cell technologies, the Kidney Precision Medicine Project (KPMP) together with the Human Biomolecular Atlas Program (HuBMAP) and the Human Cell Atlas (HCA) consortia bring us an atlas of healthy and injured cell states and niches in the human kidney. This atlas is a giant leap forward in our understanding of how cell injury leads to organ compromise.
Before we discuss the paper, let’s review some of the key advancements in single cell genomics technology and their applications in the kidney that have paved the way for this milestone:
A brief overview of single cell technology
In 2009, the world was introduced to single-cell RNA sequencing (scRNA-seq) when a group in Japan published its findings in a 4-cell mouse blastomere. At its core, scRNA-seq allows researchers to dissect specimens into their constituent cells and analyze the unique genetic signatures of each individual cell. Thanks to several technological advancements that followed, such as the use of microfluidics systems to separate cells and cell multiplexing, tens of thousands of cells can be assessed in current day scRNA-seq experiments.
Since the inception of scRNA-seq, several newer and complementary single cell technologies have emerged. In single-nucleus RNA sequencing (snRNA-seq), is particularly useful for studying tissues where the extraction of intact single cells is challenging, such as fixed or frozen kidney biopsy tissue. Single-nucleus ATAC seq (snATAC-seq) assesses DNA chromatin accessibility as a proxy of gene expression. More recent methodologies (such as SNARE-seq2 used in this paper) permit combined assessment of snRNA and snATAC in the same cell, which provides a more precise estimate of gene expression.
The aforementioned techniques provide gene expression profiles for individual cells that have been separated from their tissue niche. Spatial genomics techniques use a complementary approach: expression profiles for thousands of genes are assessed in situ in very small sections, with the latest techniques (such as Visium and Slide-seqV2 used in this paper) achieving near-cell resolution. This NephJC paper combines high-resolution single-cell and single-nucleus data with spatial transcriptomic estimates to map out cellular niches in the kidney.
What was known: kidney injury and disease at a single cell level
Single-cell technologies have yielded many new insights as it pertains to the kidney, from embryonic development processes to kidney disease and fibrosis mechanisms. This excellent review by Felix Schreibing and Rafael Kramann aptly highlights many key single cell studies that have advanced our understanding of kidney injury. These include:
Park et al. (Science, 2018): Identified a subset of intercalated cells that transformed into principal cells due to Notch pathway activity and contributed to metabolic acidosis in a mouse model of CKD. We covered this paper in our first ever single cell NephJC.
Kirita et al. (PNAS, 2020): Identified that VCAM1+ is a marker for a proximal tubular cell state that fails to repair after AKI and instead is proinflammatory and profibrotic.
Kuppe et al. (Nature, 2020): Identified the cellular origin of myofibroblasts, the cells that mediate kidney fibrosis (the final common endpoint in CKD). We covered this paper in our second single cell NephJC.
Conway et al. (JASN, 2020): Identified novel immune cell subsets involved in AKI, including Arg1+ monocytes during early injury and Mmp12+ macrophages in post-AKI repair, by building a single cell atlas of a common mouse model of AKI (unilateral ureteral obstruction).
Hansen et al. (Science Advances, 2022): A comprehensive reference atlas of the human kidney, authored by the KPMP consortium that also led the current kidney injury atlas. Founded in 2017, KPMP collects kidney biopsy specimens with a goal of building kidney tissue atlases.
With that, let’s dive into the paper!
The Study
Methods
To create a high-resolution cellular atlas of the kidney in healthy and disease states, the authors applied single-cell, single-nucleus, and spatial transcriptomics assays to adult kidney biopsy samples from 45 healthy reference donors and 48 patients with kidney disease. More data on participant characteristics are available in Supplementary Table 3. In brief, participants were evenly split between men and women, and data on self-identified race and ancestry were not available for all patients, although many self-identified as white or Black. All were adults between the ages of 20 and 80, and those with CKD did not have the etiology of their disease identified, but pathology findings, A1c, presence of albuminuria, and duration of hypertension were noted for most.
Figure 1a shows an overview of the workflow from biopsy to omics assays:
Details on how the kidneys biopsies were processed for single-nucleus assays are outlined in detail here, and the protocol for processing biopsies for single-cell assays is described here. Once the kidney tissues were dissociated into single cells or had single nuclei extracted, each single cell or single nucleus needed to have its RNA and/or chromatin accessibility evaluated separately. The general principle behind this is having each single cell or nucleus labeled with a unique barcode so that all data with that barcode can later be binned together as being from that cell. This is immensely important for demultiplexing millions (or billions) of sequencing reads in a data analysis pipeline and being able to sort which reads belong to which cells.
To process samples for single-cell outputs, the authors employed droplet-based technology through the 10X Genomics platform; and to improve scalability, they also moved to SNARE-seq2, which leverages a combinatorial indexing platform to increase the number of captured nuclei for both chromatin accessibility and transcriptomics assays.
Below is a schematic overview from 10X Genomics of their droplet-based single-cell / single-nucleus technology, and their informative brief video is available here.
Because the number of single cells or nuclei that can be processed through this droplet-based approach is limited, SNARE-seq2 (summarized in figure below), was employed to scale up assay capacity for both transcriptomics and chromatin accessibility.
As mentioned above, these methods were employed to examine transcriptomics and chromatin accessibility on a single-cell or single-nucleus level. Transcriptomic or gene expression assays capture the RNA transcripts that are present in a sample, but why look at chromatin accessibility in the same cell or nucleus? In short, identifying open chromatin regions and comparing those regions to gene expression data can potentially identify regulators of gene expression, including transcription factors whose binding sites are captured in chromatin accessibility assays. For more general information on ATAC-seq, you can read a summary here.
Some biopsy samples also underwent spatial transcriptomics, which captures gene expression profiles of many small “spots” on a tissue sample. Each “spot” can capture the gene expression of 1-10 cells that line up with its unique positional barcodes on the slide. Although currently this technology does not capture data on a single-cell level, the overall transcriptomic architecture of the tissue microenvironment can still be seen. Below is an overview of how Visium, a spatial platform from 10X Genomics, works (image from 10X Genomics).
Below is an overview of Slide-seq (v2 has improvements in RNA capture efficiency and library generation), which leverages an array of beads with unique barcodes. Because the beads are smaller than the feature spots of Visium, Slide-seq might capture data closer to the single-cell level.
Only a subset of the collected kidney biopsies underwent multiple omics assays, as summarized in Figure 1b:
After processing tissues for the omics assays, much data also needed to be demultiplexed, filtered for quality control, and analyzed in order to identify cell types, see each cell’s transcriptional and chromatin accessibility profile, and see trends across cell populations captured in the following UMAP in Figure 1c:
The authors describe in detail their bioinformatics pipeline for their analyses, which are outside the scope for a general NephJC summary but can certainly be discussed in more detail during the live chat.
With these analyses, the authors correlated clinical phenotype with kidney cell states (adaptive vs. maladaptive). They used data and biosamples from the NEPTUNE cohort (193 adult patients with proteinuria) and European cDNA Bank (ERCB, 131 patients) cohort. Leveraging bulk transcriptomics of the tubulointerstitial compartment of patient biopsies, the authors generated composite scores for genes enriched in maladaptive and adaptive cell populations seen in their single cell and single nucleus data. For primary outcome (40% loss of eGFR or development of ESKD) analyses, patients were grouped by tertiles of cell state score. Kaplan-Meier analyses were performed to calculate significance between patients of different cell state tertiles, and multivariable adjusted Cox proportional hazard analyses were performed using five different models to adjust for clinical covariates.
Results
What does the kidney look like at the single cell level, with all these different technologies, and with the inclusion of AKI and CKD biopsies?
The authors discovered 100 distinct cell populations that were annotated into 77 subclasses of epithelial, endothelial, stromal, immune, and neural cell types.
Fig 2A, B
The authors provide beautiful resolution of the glomerulus where they found glomerular endothelial cells, podocytes, parietal cells and expected marker genes like NPHS2 (nephrin for the podocyte) (Figure 2E,F). In their slide-seq2 they found macula densa (high expression of NOS1, neuronal nitric synthase), and juxtaglomerular cells (REN, renin), right beside the glomerulus as you would expect (Figure 2E). These types of genetic signatures coupled with spatial orientation allows for deeper questions about these cells both in physiological and pathological situations.
One of the new advances in this reference atlas was the characterization of cellular stress and identification of pathophysiological gene signatures. These putative states were called:
Reference cell state (ref) – think of these as quiescent cells.
Cycling (cyc) – cells that have entered the cell cycle.
Transitioning – from one nephron segment to another, like the ascending thin limb to the medullary thick ascending limb. This was visualized with slide-seq.
Adaptive (a)- successful or maladaptive repair
Degenerative (d) - damaged or stressed.
In Figure 3D, you can see examples of reference genes and those with high expression in adaptive and degenerative states. For example, in proximal tubules, healthy ones express LRP2 (the gene that encodes megalin). However, degenerate proximal tubules have very low LRP2 and adaptive proximal tubules have little LRP2 and high SPP1 (a known injury marker).
Figure 3A-D
The authors took the genetic signatures for these altered states and calculated the proportion of cells in those states between healthy control references, AKI, and CKD biopsies. As shown in Figure 3A, altered cellular states were enriched in the AKI and CKD samples, and fibroblasts and immune cells had significant associations with epithelial cells of altered states (Figure 3C,D). This makes sense because we know damaged tubules lead to factors being released that cause myofibroblast transition and recruitment of immune cells. This was nicely applied in Figure 3E,F, where the authors mapped back genetic signatures to the hematoxylin & eosin image of a fibrotic area from a CKD biopsy. Here you can see the interstitial fibrosis near dilated and damaged proximal tubules. Because the spatial resolution isn’t at the single cell, instead a spot represents many cells, they are called neighborhoods. Near that fibrosis area, there are genetic signatures like adaptive proximal tubules, adaptive fibroblasts, monocyte-derived cells, and neutrophils in the fibrosed area.
Figure 3E,F
To further uncover in situ cellular niches and injured microenvironments across kidney disease, 3D multiplexed immuno- fluorescence imaging and label-free cytometry (3DTC) was performed (Figure 4). 14 cellular niches through community detection that included expected niches like cortical epithelium (N7, and N8, Figure 4B). Niches 7 and 8 are enriched in areas of injury and predominantly consist of proximal tubules (PT, niche N7) and cortical thick ascending limbs (TAL, niche N8). A great example of the application of this technology is in Figure 4A (iii). This is niche N3 that has epithelial degeneration, fibrosis, and CD3+ T cells. Compare this to myeloid cells, that were found in niches with cortical and medullary epithelium (N6, N11), which is consistent with them being associated with adaptive rather than degenerative epithelia. Myeloid cells are also persistent in mouse ischemia-reperfusion injured kidneys. Neutrophils co-localize with both adaptive and degenerative states and may infiltrate along with T cells. Thus, the authors have identified different altered states of injured tubules within neighborhoods and they have distinct-immune-active cellular niches.
Figure 4
What about the progression and potential pathology of altered tubular epithelium?
Degenerative states are too disconnected and likely too injured to repair (Extended data Figure 9A) but adaptive (a) epithelial cells show dynamic changes from dedifferentiated to mature functional states. aTAL cells in both humans and mice showed repair trajectories leading to cTAL and mTALs (Figure 5A). In human CKD and AKI samples, aPTs also have trajectories leading to PTS1-S3 (Extended data Figure 9A). This led to the authors identifying a list of repair signatures that were conserved between the humans and mice or distinct to humans (Figure 5B)
Figure 5a, Extended Data Figure 9A, Figure 5B
A nice inclusion to this dataset was further analyses with the time course of injury and repair following renal ischemia-reperfusion-injury in male mice. At 4 hours after IRI, there are many degenerate proximal tubules (salmon colored dots) and presence of adaptive epithelial cells (yellow dots) (Extended data Figure 6E), and by 12 hours there is a loss of PTs and TALs (presumably those so badly injured that they failed to repair). Then by 2 days after injury remaining epithelial cells start cycling and dividing, as the kidney continues to heal over the 6-week period of study.
Extended data Figure 6E.
Epithelial repair signatures include many growth factors, but we know these are also involved in maladaptive repair, fibrosis, and inflammation (TGFβ, Wnt, JAK/STAT, etc, Figure 5C). Chromatin accessibility of transcription binding sites were mapped along with their gene expression signatures for early (blue), mid (yellow), late(red) repair states or reference (pink and black) states. We can use TGFβ as an example. It signals through the SMADs and as you can see in Figure 5C, SMAD2 and SMAD3 expression is greater in mid repair (bigger dot, darker blue), and SMAD DNA binding sites are more accessible in mid repair too (bigger dot, dark purple).
Figure 5C
But adaptive states can be maladaptive too?
We’ve known for quite some time that many pathways activated after injury and early in repair, if sustained, can become maladaptive. Classic examples are inflammation and fibrosis which are examples of this “double edged sword” [Black et al. 2019]. In this study, the authors looked at this in detail with their statistically significant association between the aTAL cell signatures early in repair that significantly associated with disease progression in the Nephrotic Syndrome Study Network (NEPTUNE) cohort of 193 patients. Genome-wide transcriptome analysis was performed on the NEPTUNE biopsies, and composite scores for genes enriched in aPT, aTAL, an aStr (adaptive stromal cells) and then the results were binned according to the degree of cell state score by tertile. In Figure 6E, these are the unadjusted Kaplan-Meier curves by cell state for composite end-stage renal disease (or 40% drop in eGFR from time of biopsy) and significant associations were observed with the aPT, aTAL, aStr but not degenerative states. Similar findings were also observed in patients with diabetes, hypertension, and focal segmental glomerulosclerosis from the European Renal cDNA Bank cohort. Thus, overtime after the biopsies, those individuals in the highest tertile for cell signatures in the aPT, aTAL, or aStr had the greatest probability of ESRD.
Figure 6E, Extended Data Figure 12B
Discussion
This is a large amount of human and rodent data, integration of novel technologies and clinical outcomes and provides the first comprehensive atlas of healthy and injured kidney states, neighborhoods, and niches in humans. The authors have defined and validated many genes associated with degenerative, adaptive, and cycling states, and these genes and gene networks can be applied to other kidney diseases. For example, PROM1 (CD133) encodes a pentaspan transmembrane glycoprotein that functions to maintain stem cell-like properties by suppressing differentiation. The PROM1 RNA is expressed in the damaged PTs and TALs early during repair, and PROM1 protein is detected in damaged tubules. There are preclinical studies testing if PROM1-positive extracellular vesicles could be a therapeutic approach to preserve kidney function in CKD models [Miyasaki et al. 2022].
There are limitations to these studies including the intrinsic heterogeneity of biopsies, and it wasn’t possible to run the same sample through all technologies. But given that many of the top genes were also found in the preclinical models, it is amazing how the response to injury in the kidney is well conserved. As technology advances, and we move from single cell/nucleus RNA to single cell proteomics, we will continue to define the pathways involved in kidney injury, repair, and maladaptive states. With these data hopefully new therapeutic approaches can be developed to stop kidney function decline and maybe even repair injured and diseased areas of the kidney.
Summary by
NephJC Basic Science Editors
Kelly Hyndman
University of Alabama at Birmingham
Jennie Lin
Northwestern University, Chicago
and Caitlyn Vlasschaert
Queen’s University, Kingston, Canada